On October 13, 2026, Google DeepMind officially released Gemini 4 Argon. This is Google's most advanced frontier model to date, announced personally by Koray Kavukcuoglu, Senior Vice President and Chief AI Architect at Google DeepMind, with API access immediately available to developers and enterprises.
Argon ranked first on the Vals Index industry benchmark with a score of 68.90%. According to vals.ai data, second-place Claude Sonnet 5.5 scored 67.04%, followed closely by Claude Opus 5.5 at 66.97%. Its lead is less than 2 percentage points.
Pricing Is the Real Strategic Weapon
On its first day of release, Argon uses introductory pricing: $2 per million input tokens, $10 per million output tokens, with an additional 95% discount on cached input.
For comparison with similar flagship models: GPT-6 and Claude Fable are both priced at $10 per million input tokens and $50 per million output tokens. For the same output volume, Argon's bill is only one-fifth that of its competitors.
This price has an expiration date—the introductory pricing will double on December 1, 2026. Google's intent in setting this time window is clear: exchange low prices for ecosystem lock-in and establish switching costs before users complete workflow migration.
Vals Index: A New Yardstick
The Vals Index on which Argon took first place is not Google's own test. Vals AI is backed by a16z; its evaluation framework focuses on the professional domains with the highest economic value, such as finance, law, tax, and programming. Each domain is weighted according to its share of U.S. GDP, and the test question bank is not publicly disclosed.
Within this framework, Argon's performance is not an across-the-board rout. According to comparative evaluations:
- Legal work (Harvey Legal Agent benchmark): Argon leads by a wide margin at 19.6%, compared with 5.4% for GPT-6 Astra
- Software engineering (DeepSWE v1.1): Argon 77.9%, Claude Opus 5.5 74.2%, GPT-6 Astra 74.1%, a lead of about 4 percentage points
- Terminal interactive programming (Terminal-Bench 4.0): Argon about 57%, trailing Claude Opus 5.5 at 60%+
The large lead in legal work is the key support for Argon's Vals Index win. Its structural advantage is concentrated in legal tasks that rely heavily on precise reasoning and cross-referencing across multiple documents.
Chrome Codebase Refactoring: The Most Convincing Real-World Data
The Chromium codebase has 40 million lines in total, and one legacy folder holding browser features contains about 6 million lines. The Chrome team needed to separate core web-processing logic from this historical burden. After using Argon, months of work were compressed into weeks.
Inside Google, teams using Argon achieved an increase of more than 35% in agentic code commits, while the code rollback rate fell by nearly one-third.
There is also quantitative data on the research side: Argon helped increase the molecular structure prediction speed of the xTB quantum chemistry software package by 2.3 times; on an A100 GPU, it raised Gemma 4 inference throughput from about 37,900 tokens/second to 108,000+ tokens/second.
Why Google Needs Argon Now
Over the past two years, OpenAI's and Anthropic's flagship models have dominated enterprise purchasing mindshare: the o series focuses on reasoning, while Opus focuses on high-quality long-context tasks. Google's Flash series wins on speed and cost, but it has always been absent from the category perception of "the hardest tasks."
Argon is named after the chemical stability of the noble gas, a metaphor for reliability in complex, high-pressure environments. Its official positioning, "built to solve your hardest problems," directly targets the purchasing criterion in enterprise decision-makers' minds: "which model can handle the toughest tasks."
The $2/$10 introductory price is not the result of cost pressure but a deliberate compression of profit in exchange for a ticket to entry. Google's calculation is clear: getting enterprises to embed Argon in their workflows is more effective than any marketing campaign—when the price doubles on December 1, customers that have already completed migration will face a cost trade-off between continuing to use it and switching again.
Independent Judgment: A Successful Entry, Not a Wholesale Disruption
Based on the available data, Argon is an effective entry by Google into the enterprise AI market, but it is not a complete disruption of the industry landscape.
The Vals Index lead is real but limited in margin. The cliff-like advantage in legal work (19.6% vs. 5.4%) is a highlight, but its lag in terminal interactive programming reveals a structural tilt in the model's capabilities: Argon is better at document-processing tasks that require deep reading and precise output, rather than developer workflows that require rapid iteration and immediate feedback.
The pricing window is a critical testing period before the price doubles on December 1. Within this window, legal tech, financial compliance, and large-codebase management teams are the early-adoption areas most worth watching.
The Chrome team's refactoring case is more significant than all the benchmark numbers. It proves that in real engineering tasks with clear goals, rule constraints, and verification mechanisms, Argon can operate autonomously and deliver trustworthy results. Google used its own most complex codebase as an endorsement.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接